Refining functional outcome prediction for proximal femoral fractures: a hierarchical machine learning approach

This study aimed to develop a refined machine learning model to predict the functional prognosis of patients with proximal femoral fractures (PFF). Given the difficulty of directly classifying all nine ADL categories, we adopted a hierarchical classification (HC) approach that decomposes the nine-class prediction problem into stepwise classification tasks. Using data from the Japan Association of Rehabilitation Database, we analyzed 2,959 patients with PFF from an initial cohort of 3,088 cases, after excluding 129 cases with missing discharge ADL outcomes. We sought to optimize rehabilitation planning tailored to individuals. The target variable was the 9-class functional prognosis at discharge. The HC framework first classified patients into five major ADL categories (Normal, J, A, B, and C), followed by binary subclassification within the J, A, B, and C categories. We evaluated 16 machine learning models and one deep learning model (TabNet). Model performance was assessed using accuracy, and interpretability was evaluated via SHapley Additive explanations (SHAP). The final HC model achieved an accuracy of 0.345 in the internal holdout cohort. This refined model has the potential to enhance preoperative assessment and facilitate more effective early postoperative recovery.

Authors

Institutions

Publication Details

Journal
Scientific Reports
Published
2026-11-14
DOI
https://doi.org/10.1038/s41598-026-74986-3
Primary Topic
Hip and Femur Fractures
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Refining functional outcome prediction for proximal femoral fractures: a hierarchical machine learning approach

Seiji Ohtori, Sumihisa Orita, Satoshi Maki, Takashi Hozumi et al.
Scientific Reports
Hip and Femur Fractures
article

Refining functional outcome prediction for proximal femoral fractures: a hierarchical machine learning approach

Seiji Ohtori, Sumihisa Orita, Satoshi Maki, Takashi Hozumi, Yuya Kawarai, Yasuhiro Shiga, Yawara Eguchi, Jun Nakamura, Issei Tanaka, Masahiro Inoue, Kazuhide Inage
article en

Abstract

This study aimed to develop a refined machine learning model to predict the functional prognosis of patients with proximal femoral fractures (PFF). Given the difficulty of directly classifying all nine ADL categories, we adopted a hierarchical classification (HC) approach that decomposes the nine-class prediction problem into stepwise classification tasks. Using data from the Japan Association of Rehabilitation Database, we analyzed 2,959 patients with PFF from an initial cohort of 3,088 cases, after excluding 129 cases with missing discharge ADL outcomes. We sought to optimize rehabilitation planning tailored to individuals. The target variable was the 9-class functional prognosis at discharge. The HC framework first classified patients into five major ADL categories (Normal, J, A, B, and C), followed by binary subclassification within the J, A, B, and C categories. We evaluated 16 machine learning models and one deep learning model (TabNet). Model performance was assessed using accuracy, and interpretability was evaluated via SHapley Additive explanations (SHAP). The final HC model achieved an accuracy of 0.345 in the internal holdout cohort. This refined model has the potential to enhance preoperative assessment and facilitate more effective early postoperative recovery.

Scientific Reports
Chiba University (JP), Center for Frontier Medical Engineering, Chiba University (JP)
Hip and Femur Fractures
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon